Reproduce “Learning Interpretable Hierarchical Dynamical Systems Models from Time Series Data” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Manuel Brenner, Elias Weber, Georgia Koppe, Daniel Durstewitz

A hierarchical framework learns interpretable generative models for underlying system dynamics from multi-domain time series, improving Dynamical Systems Reconstruction (DSR) performance, especially with short individual series, and revealing low-dimensional feature spaces that directly map to system control parameters. This method achieved superior agreement with ground truth dynamics compared to non-hierarchical and existing transfer learning approaches.

Key insights

  • Leveraging group-level information through a hierarchical structure significantly enhances DSR quality and robustness, particularly for short and sparse time series originating from diverse dynamical regimes.
  • The framework intrinsically learns highly interpretable, low-dimensional subject-level feature spaces that often exhibit a clear, frequently linear relationship with the underlying system's control parameters.
  • The DSR-derived feature spaces offer a principled basis for classifying time series data by dynamically distinct regimes, outperforming conventional time series feature extraction methods in unsupervised settings.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.